Date of Award
Doctor of Philosophy (PhD)
Revenue for data centers today is highly dependent on the satisfaction of their enterprise customers. These customers often require various features to migrate their businesses and operations to the cloud. Thus, clouds today introduce new features at a swift pace to onboard new customers and to meet the needs of existing ones. This pace of innovation continues to grow on super linearly, e.g., Amazon deployed 1400 new features in 2017. However, such a rapid pace of evolution adds complexities both for users and the cloud. Clouds struggle to keep up with the deployment speed, and users struggle to learn which features they need and how to use them. The pace of these evolutions has brought us to a tipping point: we can no longer use rules of thumb to deploy new features, and customers need help to identify which features they need. We have built two systems: Janus and Cherrypick, to address these problems. Janus helps data center operators roll out new changes to the data center network. It automatically adapts to the data center topology, routing, traffic, and failure settings. The system reduces the risk of new deployments for network operators as they can now pick deployment strategies which are less likely to impact users’ performance. Cherrypick finds near-optimal cloud configurations for big data analytics. It adapts allows users to search through the new machine types the clouds are constantly introducing and find ones with a near-optimal performance that meets their budget. Cherrypick can adapt to new big-data frameworks and applications as well as the new machine types the clouds are constantly introducing. As the pace of cloud innovations increases, it is critical to have tools that allow operators to deploy new changes as well as those that would enable users to adapt to achieve good performance at low cost. The tools and algorithms discussed in this thesis help accomplish these goals.
Alipourfard, Omid, "Change Management Systems for Seamless Evolution in Data Centers" (2021). Yale Graduate School of Arts and Sciences Dissertations. 6.